diff options
| -rw-r--r-- | nixos/modules/services/misc/ollama.nix | 11 | ||||
| -rw-r--r-- | pkgs/tools/misc/ollama/default.nix | 32 |
2 files changed, 30 insertions, 13 deletions
diff --git a/nixos/modules/services/misc/ollama.nix b/nixos/modules/services/misc/ollama.nix index 30c2b26d8322e..b2c3de09015b9 100644 --- a/nixos/modules/services/misc/ollama.nix +++ b/nixos/modules/services/misc/ollama.nix @@ -40,14 +40,19 @@ in ''; }; acceleration = lib.mkOption { - type = types.nullOr (types.enum [ "rocm" "cuda" ]); + type = types.nullOr (types.enum [ false "rocm" "cuda" ]); default = null; example = "rocm"; description = '' What interface to use for hardware acceleration. - - `rocm`: supported by modern AMD GPUs - - `cuda`: supported by modern NVIDIA GPUs + - `null`: default behavior + if `nixpkgs.config.rocmSupport` is enabled, uses `"rocm"` + if `nixpkgs.config.cudaSupport` is enabled, uses `"cuda"` + otherwise defaults to `false` + - `false`: disable GPU, only use CPU + - `"rocm"`: supported by most modern AMD GPUs + - `"cuda"`: supported by most modern NVIDIA GPUs ''; }; environmentVariables = lib.mkOption { diff --git a/pkgs/tools/misc/ollama/default.nix b/pkgs/tools/misc/ollama/default.nix index 74bd213b043f5..a377d200fbb8e 100644 --- a/pkgs/tools/misc/ollama/default.nix +++ b/pkgs/tools/misc/ollama/default.nix @@ -19,23 +19,24 @@ , linuxPackages , darwin - # one of `[ null "rocm" "cuda" ]` -, acceleration ? null - , testers , ollama + +, config + # one of `[ null false "rocm" "cuda" ]` +, acceleration ? null }: let pname = "ollama"; # don't forget to invalidate all hashes each update - version = "0.1.30"; + version = "0.1.31"; src = fetchFromGitHub { owner = "jmorganca"; repo = "ollama"; rev = "v${version}"; - hash = "sha256-+cdYT5NUf00Rx0fpCvWUNg4gi+PAOmZVDUdB3omibm0="; + hash = "sha256-Ip1zrhgGpeYo2zsN206/x+tcG/bmPJAq4zGatqsucaw="; fetchSubmodules = true; }; vendorHash = "sha256-Lj7CBvS51RqF63c01cOCgY7BCQeCKGu794qzb/S80C0="; @@ -55,13 +56,24 @@ let }; - validAccel = lib.assertOneOf "ollama.acceleration" acceleration [ null "rocm" "cuda" ]; - - warnIfNotLinux = api: (lib.warnIfNot stdenv.isLinux + accelIsValid = builtins.elem acceleration [ null false "rocm" "cuda" ]; + validateFallback = lib.warnIf (config.rocmSupport && config.cudaSupport) + (lib.concatStrings [ + "both `nixpkgs.config.rocmSupport` and `nixpkgs.config.cudaSupport` are enabled, " + "but they are mutually exclusive; falling back to cpu" + ]) + (!(config.rocmSupport && config.cudaSupport)); + validateLinux = api: (lib.warnIfNot stdenv.isLinux "building ollama with `${api}` is only supported on linux; falling back to cpu" stdenv.isLinux); - enableRocm = validAccel && (acceleration == "rocm") && (warnIfNotLinux "rocm"); - enableCuda = validAccel && (acceleration == "cuda") && (warnIfNotLinux "cuda"); + shouldEnable = assert accelIsValid; + mode: fallback: + ((acceleration == mode) + || (fallback && acceleration == null && validateFallback)) + && (validateLinux mode); + + enableRocm = shouldEnable "rocm" config.rocmSupport; + enableCuda = shouldEnable "cuda" config.cudaSupport; rocmClang = linkFarm "rocm-clang" { |